Jon Risch's Web Site
A NEW CLASS OF IN-BAND MULTITONE TEST SIGNALS
Jon M. Risch, Peavey Electronics Corp., Meridian, MS. USA
Copyright 1998
Presented at the 105th Convention of the Audio Engineering Society
1998 September 26-29th, San Francisco, California as preprint #4803
CONT'D
3.0 Experimental Results
Experimental multitone test signals were generated using a PC based waveform
generator operating in the digital domain, and transferring the final product
to CD-R for playback and use.
It was found that some CD players were unable to handle output to full
0 dBFS, with the result a much higher level of distortion. It was
as if some component was either clipping or compressing the signal, generating
harmonics, intermodulation and crossmodulation products all throughout the
spectrum. Accordingly, it was found that reducing the level of the digitally
generated recorded signal by 2 dB was enough for most players to avoid this
problem. In order to avoid any vestige of amplitude related compression
or clipping, the signals were generated and recorded with peak signal levels
below -3 dB. Some players still had a problem with these peak levels,
but most did not. The author suspect’s the digital filtering within
the CD player may be the cause.
3.1 Revisions
The original Phi12 stimulus had some distortion product cover up, due to
some of the intermediate tones intermodulating with each other, and generating
the same frequency. Once this was discovered, a revised version that
used different multipliers for each of the 6 base tones was devised.
While this did include some arbitrary multiplier choices, and could conceivably
be improved, it was found useful as a more complex stimulus than the 6 tone
version.
Accordingly, from the base frequency of 100 Hz, a multiplier of 1.22 was
used, resulting in 122 Hz. From 261.8 Hz, a multiplier of 1.33 was
used. resulting in 348.2 Hz, from 685.4 Hz, a multiplier of 1.44 resulting
987.0 Hz, from 1794 Hz a multiplier of 1.6 resulting in 2870.4 Hz, from 4697.9
Hz, a multiplier of 1.44 resulting in 6765.0 Hz, from 12299 Hz, a multiplier
of 1.33, resulting in 16358 Hz. None of these has obvious intermodulation
or crossmodulation products that occur at exactly the same frequency.
Most distortion products are far enough apart for a spectrum analyzer to differentiate
between the different products. Figure 3 is the FFT spectrum of this
signal, from a CD player through a mixer.
4.0 Further Test Signal Refinements
Other changes and refinements were considered to reduce or alleviate some
of the limitations of the signal.
4.1 Possible Crest Factor Reduction
Proper phasing of the various signal tones can provide some relief, but
a DUT that had significant amplitude changes, roll-off or phase shifts would
destroy the carefully orchestrated phase relationships of the original test
signal. It is recommended that full allowance be made for the worst
case crest factor of the multiple tones. Several references regarding
such phase manipulation are provided. (4)
4.2 Split Octave Bands - High Low
After working experimentally with some of the variations on the signals,
it was determined that more tones or a higher density of tone spacing might
be desirable to highlight intermodulation and cross-modulation distortions,
and provide more distortion products across the audio band. It was
also determined that limiting the total number of individual tones would help
preserve dynamic range.
To raise the tone interval density without increasing the number of tones
to a high number, isolated octave bands of tones was considered. A
suitably non-overlaying multiplier was desired, so a multiplier of one plus
one tenth the original full audio band multipliers were investigated.
Scatter diagrams were used to evaluate the suitability of various octave band
combinations with a reasonable total number of tones, that would provide a
wide range of harmonic, intermodulation, and crossmodulation products.
One of the concerns was to limit the spread of each group or band of signals
to an octave or less, so that the second harmonics of the band would not be
covered up by the primary tones.
A pair of split octave bands was found to provide an efficient and useful
multitone test signal without an excessive number of individual tones.
An octave starting at 100 Hz, in conjunction with an octave starting at 5
kHz, with both containing five tones, and with spacing multipliers of 1.1618.
This results in frequencies of 100, 116.18, 134.98, 156.80 and 182.19 Hz
for the low band, and of 5000, 5809, 6748.9, 7840.9, and 9109.6 Hz.
This test signal proved particularly efficacious at ferreting out intermodulation
and crossmodulation in two-way loudspeakers. Figure 4 depicts the
FFT spectrum of this signal, from a CD player through a mixer.
Note that this is similar in spirit to the third signal proposed by Sokolich
and Jensen, but not identical. Much larger areas are left open for
distortion products to manifest and be readily observed.
Other split band combinations were investigated, such as:
100 -182 Hz plus 10 - 18.2 kHz, with frequencies of 100, 116.18, 134.98,
156.80 and 182.19 Hz for the low band, and 10000, 11618, 13498, 15682, and
18219 Hz for the high band. Figure 5.
100 -182 Hz plus 1.11 - 2.02 kHz, with frequencies of 100, 116.18,
134.98, 156.80 and 182.19 Hz for the low band, and 1109.0, 1288.4, 1496.9,
1739.1, 2020.5 Hz for the high band. Figure 6.
261 - 477 Hz plus 1.11 - 2.02 kHz, with frequencies of 261.80, 304.16,
353.38, 410.56, and 476.99 Hz for the low band, and 1109.0, 1288.4, 1496.9,
1739.1, 2020.5 Hz for the high band. Figure 7.
261 - 477 Hz plus 10 - 18.2 kHz, with frequencies of 261.80, 304.16, 353.38,
410.56, and 476.99 Hz for the low band, and 10000, 11618, 13498, 15682,
and 18219 Hz for the high band. Figure 8.
685 - 1249 Hz plus 1.79 - 3.27 kHz, with frequencies of 685.41, 796.31,
925.16, 1074.9, and 1248.8 Hz for the low band, and 1794.4, 2084.8, 2422.1,
2814.0, and 3269.3 Hz for the high band. Figure 9.
4.3 Revised Split Octave Band Signal
It proves all too easy to fall into the same trap as before. In selecting
even starting frequencies, such as 100 Hz for the low band, and 10 kHz for
the high band, some of the harmonics, intermodulation and crossmodulation
products line up again, covering up the source of the tones responsible
for the distortion.
After trying a different start frequency that was not an even multiple
of the other start frequency for the high band or the low band, it was realized
that by using the same multiplier for each band, the top band of frequencies
would divide into the bottom band and have the same crossmodulation product.
This leads to the need for a different multiplier for the separate bands.
After checking for this, a slightly different multiplier for one of the
bands was used, 1.2618 vs. 1.1618.
Use of the new 1.2618 multiplier results in the high frequency band extending
for a bit more than an octave, but the author considers this to be completely
arbitrary and a minor issue. If a band of closely spaced multitone
frequencies extends past an octave, some potential exists for cover-up of
harmonics. For the higher frequency bands where this was done, the resolution
is sufficient to allow determination of the 2nd harmonics without difficulty.
This is why the low frequency band uses the originally derived multiplier.
At this point, a spread sheet was developed for use in evaluating the resultant
distortion products that would occur with various primary test signal tones
in a spectral type test signal.
Figure 10 is a copy of the spread sheet for the Phi6 spectral signal, Figure
11 is a scatter diagram of the primary tones for the Phi6 spectral, harmonics
(up to the tenth), and intermodulation and crossmodulation products.
Figure 12 is the spread sheet for the Phi12 spectral signal, and Figure 13
the scatter diagram. Figure 14 is the spread sheet for the Phi12r (revised),
and Figure 15 the scatter diagram. Figure 16 is the spread sheet for
the first split band spectral, the one with bands at 100 -182 Hz plus 5 -
9.1 kHz.
Experimental results for the first set of split band spectral test signals
reveal that the pairings with the low band at 261 to 477 Hz did not show
any additional distortion products than the ones with the low band from 100
to 182 Hz. They also do not show low frequency second harmonics as
readily. Also, the pairings that had the high band in the last octave
do not show any high frequency second harmonics, or intermodulation or crossmodulation
products on the high side before the audio band was exceeded. In order
to address this, revised split band signals were reduced to a
set of low and high band signals, and a set of low and middle band signals.
The revised split band spectral’s now use frequencies at: 100 -182
Hz plus 4.7 - 11.9 kHz, with frequencies of 100, 116.18, 134.98, 156.80 and
182.19 Hz for the low band, and 4697.9, 5927.8, 7479.7, 9437.9 and 11909 Hz
for the high band for the low/high band pairing, and called a Phi Low-High
split band spectral. Shown in Figure 17.
The low/middle revised split band now uses frequencies at: 100, 116.18,
134.98, 156.80 and 182.19 Hz for the low band, and 986.99, 1245.4, 1571.4,
1982.8 and 2502.0 Hz, and I call it a Phi Low-Middle split band spectral.
Figure 18.
4.4 Phi Tri-Band Spectral Signal
Further work with these split band spectral signals resulted in the development
of a split band signal with components at low, middle AND higher frequencies,
which left two large open areas for distortion products to show up.
The multipliers have to be different for each band. The total
number of tones needs to be kept down, so three clusters of four frequencies
was used, for a total of 12 separate tones. Multipliers of 1.1618, 1.2618
and 1.12618 were used for the three bands respectively. The frequencies
that result are: 100, 116.18, 134.98, and 156.80 Hz for the low frequency
band, 986.99, 1245.4, 1571.4, and 1982.8 Hz for the middle band, and 6764.9,
7618.5, 8579.8 and 9662.5 Hz for the high frequency band. I refer to
this one as a Phi tri-band spectral, shown in Figure 19.
The frequency ranges were chosen based on earlier work with split band
multitone’s, and what the author felt would be good ranges for excitation
frequencies. Each band has an empty space of approximately two
octaves in-between, the low band two octaves below, and the high band
an octave above (within the audio band). The start frequency of each
band is based on the revised Phi12 (Phi12r) spectral contamination test signal.
The underlying premise is that none of the frequencies within each band,
including the start frequency, would encourage distortion products to be
covered up by other products, or by the primary tones.
Cont'd
Link to 3rd
web page on this paper
Copyright Notice
All website content Copyright Jon M. Risch, 1997 - 2004
All applicable copyright laws apply, all rights reserved, except transmission
by USENET and like facilities granted. Any use or inclusion in print
or other media are specifically prohibited. The informational content
is not warrantied in any way or form, and any use of said content are at
the reader's own risk, the author shall not be held responsible in any way
for any damages or injuries arising from the content of this web site.
Common safety practices are encouraged at all times. Do not fold, spindle
or mutilate.
----> E-MAIL ME ! <-----
Back to home page